STNAGNN: Data-driven Spatio-temporal Brain Connectivity beyond FC

2024-06-17

AI Summary

fMRI-based ROI connectivity modeling faces three key challenges: high noise susceptibility, the undirected and non-causal nature of functional connectivity (FC), and low signal-to-noise ratio alongside oversimplification in sliding-window dynamic modeling. To address these, we propose a data-driven Spatio-Temporal Node-Attention Graph Neural Network (ST-NA-GNN). Our method is the first to jointly integrate sparse prior FC with dense adaptive spatio-temporal connections, incorporating a node-level spatio-temporal attention mechanism to learn dynamic, directed, and nonlinear brain connectivity. By constructing hybrid sparse-dense graphs and performing sliding-window spatio-temporal modeling, ST-NA-GNN effectively suppresses noise while preserving subject-specific patterns. Evaluated on multiple task-based and resting-state fMRI datasets, it achieves 3.2–5.8% improvements in classification accuracy. Visualization confirms that the learned connectivity patterns exhibit both neurobiological interpretability and strong subject discriminability.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Mining of Spatial, Temporal or Spatio-Temporal Data

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

In recent years, graph neural networks (GNNs) have been widely applied in the analysis of brain fMRI, yet defining the connectivity between ROIs remains a challenge in noisy fMRI data. Among all approaches, Functional Connectome (FC) is the most popular method. Computed by the correlation coefficients between ROI time series, FC is a powerful and computationally efficient way to estimate ROI connectivity. However, it is well known for neglecting structural connections and causality in ROI interactions. Also, FC becomes much more noisy in the short spatio-temporal sliding-window subsequences of fMRI. Effective Connectome (EC) is proposed as a directional alternative, but is difficult to accurately estimate. Furthermore, for optimal GNN performance, usually only a small percentage of the strongest connections are selected as sparse edges, resulting in oversimplification of complex brain connections. To tackle these challenges, we propose the Spatio-Temporal Node Attention Graph Neural Network (STNAGNN) as a data-driven alternative that combines sparse predefined FC with dense data-driven spatio-temporal connections, allowing for flexible and spatio-temporal learning of ROI interaction patterns.

Problem

• Defining noisy fMRI ROI connectivity beyond Functional Connectome
• Addressing neglect of structural connections and causality in FC
• Improving sparse edge oversimplification in GNN brain analysis

Innovation

• Combines sparse FC with dense data-driven connections
• Uses spatio-temporal node attention GNN
• Learns flexible ROI interaction patterns